전체 이미지를 대상으로 $\theta$를 바꿔가며 accuracy 변화를 확인하자

import 

import torch 
from fastai.vision.all import *
import cv2

data 

path=untar_data(URLs.PETS)/'images'

path

Path('/home/khy/.fastai/data/oxford-iiit-pet/images')
files=get_image_files(path)

def label_func(f):
    if f[0].isupper():
        return 'cat' 
    else: 
        return 'dog' 

dls=ImageDataLoaders.from_name_func(path,files,label_func,item_tfms=Resize(512))

learn 

lrnr=cnn_learner(dls,resnet34,metrics=error_rate)
lrnr.fine_tune(1)
epoch train_loss valid_loss error_rate time
0 0.152202 0.023401 0.007442 01:10
epoch train_loss valid_loss error_rate time
0 0.054687 0.010879 0.004736 01:25 
net1=lrnr.model[0]
net2=lrnr.model[1] 

net2 = torch.nn.Sequential(
    torch.nn.AdaptiveAvgPool2d(output_size=1), 
    torch.nn.Flatten(),
    torch.nn.Linear(512,out_features=2,bias=False))

net=torch.nn.Sequential(net1,net2)

lrnr2=Learner(dls,net,metrics=accuracy) 

lrnr2.fine_tune(15)
epoch train_loss valid_loss accuracy time
0 0.290503 0.831334 0.808525 01:24
epoch train_loss valid_loss accuracy time
0 0.125157 0.083141 0.968877 01:24
1 0.098165 0.124922 0.950609 01:24
2 0.095035 0.172816 0.925575 01:24
3 0.107872 0.172295 0.934371 01:24
4 0.094506 0.202725 0.937754 01:24
5 0.077550 0.111983 0.948579 01:24
6 0.077960 0.432051 0.845737 01:24
7 0.054796 0.079238 0.970907 01:24
8 0.046034 0.099135 0.970230 01:24
9 0.034132 0.064647 0.979702 01:24
10 0.023779 0.074031 0.980379 01:24
11 0.013524 0.056572 0.981732 01:24
12 0.009503 0.062131 0.983762 01:24
13 0.006159 0.051805 0.984438 01:24
14 0.005125 0.057023 0.983762 01:24 
interp = ClassificationInterpretation.from_learner(lrnr2)
interp.plot_confusion_matrix()

interp = Interpretation.from_learner(lrnr2)

interp.plot_top_losses(16)

$\theta=0.1$

1st 

files

(#7390) [Path('/home/khy/.fastai/data/oxford-iiit-pet/images/boxer_128.jpg'),Path('/home/khy/.fastai/data/oxford-iiit-pet/images/Sphynx_142.jpg'),Path('/home/khy/.fastai/data/oxford-iiit-pet/images/British_Shorthair_203.jpg'),Path('/home/khy/.fastai/data/oxford-iiit-pet/images/Ragdoll_142.jpg'),Path('/home/khy/.fastai/data/oxford-iiit-pet/images/Persian_272.jpg'),Path('/home/khy/.fastai/data/oxford-iiit-pet/images/Bombay_200.jpg'),Path('/home/khy/.fastai/data/oxford-iiit-pet/images/shiba_inu_103.jpg'),Path('/home/khy/.fastai/data/oxford-iiit-pet/images/chihuahua_142.jpg'),Path('/home/khy/.fastai/data/oxford-iiit-pet/images/scottish_terrier_156.jpg'),Path('/home/khy/.fastai/data/oxford-iiit-pet/images/basset_hound_163.jpg')...]
fig, ax = plt.subplots(5,5) 
k=0 
for i in range(5):
    for j in range(5): 
        x, = first(dls.test_dl([PILImage.create(get_image_files(path)[k])]))
        camimg = torch.einsum('ij,jkl -> ikl', net2[2].weight, net1(x).squeeze())
        a,b = net(x).tolist()[0]
        catprob, dogprob = np.exp(a)/ (np.exp(a)+np.exp(b)) ,  np.exp(b)/ (np.exp(a)+np.exp(b)) 
        if catprob>dogprob: 
            test=camimg[0]-torch.min(camimg[0])
            A1=torch.exp(-0.1*test)
            X1=np.array(A1.to("cpu").detach(),dtype=np.float32)
            Y1=torch.Tensor(cv2.resize(X1,(512,512),interpolation=cv2.INTER_LINEAR))
            x1=x.squeeze().to('cpu')*Y1-torch.min(x.squeeze().to('cpu')*Y1)
            (x1*0.35).squeeze().show(ax=ax[i][j])
            ax[i][j].set_title("cat(%s)" % catprob.round(5))
        else: 
            test=camimg[1]-torch.min(camimg[1])
            A1=torch.exp(-0.1*test)
            X1=np.array(A1.to("cpu").detach(),dtype=np.float32)
            Y1=torch.Tensor(cv2.resize(X1,(512,512),interpolation=cv2.INTER_LINEAR))
            x1=x.squeeze().to('cpu')*Y1-torch.min(x.squeeze().to('cpu')*Y1)
            (x1*0.35).squeeze().show(ax=ax[i][j])
            ax[i][j].set_title("dog(%s)" % dogprob.round(5))
        k=k+1 
fig.set_figwidth(16)            
fig.set_figheight(16)
fig.tight_layout()

Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).

img = PILImage.create(get_image_files(path)[3])
img
x, = first(dls.test_dl([img]))

fig, (ax1,ax2,ax3) = plt.subplots(1,3) 
# 
dls.train.decode((x,))[0].squeeze().show(ax=ax1)
ax1.set_title("Input image")
# 
dls.train.decode((x,))[0].squeeze().show(ax=ax2)
ax2.imshow(camimg[0].to("cpu").detach(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax2.set_title("CAT PART")
#
dls.train.decode((x,))[0].squeeze().show(ax=ax3)
ax3.imshow(camimg[1].to("cpu").detach(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax3.set_title("DOG PART")
#
fig.set_figwidth(12)            
fig.set_figheight(12)
fig.tight_layout()

fig, ax = plt.subplots(5,5) 
k=0 
for i in range(5):
    for j in range(5): 
        x, = first(dls.test_dl([PILImage.create(get_image_files(path)[k])]))
        camimg = torch.einsum('ij,jkl -> ikl', net2[2].weight, net1(x).squeeze())
        a,b = net(x).tolist()[0]
        catprob, dogprob = np.exp(a)/ (np.exp(a)+np.exp(b)) ,  np.exp(b)/ (np.exp(a)+np.exp(b)) 
        if catprob>dogprob: 
            test=camimg[0]-torch.min(camimg[0])
            A1=torch.exp(-0.015*test)
            X1=np.array(A1.to("cpu").detach(),dtype=np.float32)
            Y1=torch.Tensor(cv2.resize(X1,(512,512),interpolation=cv2.INTER_LINEAR))
            x1=x.squeeze().to('cpu')*Y1-torch.min(x.squeeze().to('cpu')*Y1)
            (x1*0.35).squeeze().show(ax=ax[i][j])
            ax[i][j].set_title("cat(%s)" % catprob.round(5))
        else: 
            test=camimg[1]-torch.min(camimg[1])
            A1=torch.exp(-0.015*test)
            X1=np.array(A1.to("cpu").detach(),dtype=np.float32)
            Y1=torch.Tensor(cv2.resize(X1,(512,512),interpolation=cv2.INTER_LINEAR))
            x1=x.squeeze().to('cpu')*Y1-torch.min(x.squeeze().to('cpu')*Y1)
            (x1*0.35).squeeze().show(ax=ax[i][j])
            ax[i][j].set_title("dog(%s)" % dogprob.round(5))
        k=k+1 
fig.set_figwidth(16)            
fig.set_figheight(16)
fig.tight_layout()

Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).